The Lasso is an optimization problem devoted to find a sparse representation of some signal with respect to some predefined dictionary. We propose an original and computationally efficient method to solve the Lasso problem. Our method rests upon the idea of dynamic screening test and allows us to accelerate first-order algorithms. At each iteration we take advantage of the computation done for the optimization process to reduce the size of the dictionary by discarding elements that will surely not enter the sparse representation. As this screening step is not expensive, the computational cost of the algorithm using the dynamically screened dictionary is cheaper than the standard algorithm. Numerical simulations on synthetic and real data support the relevance of our approach.